>  | Are you sure about the following?
>  | 
>  | @@ -82,10 +82,10 @@
>  |         void *p;
>  |         int i, new_count;
>  | 
>  | -       if (state < fa->state_count)
>  | +       if (++state < fa->state_count)
>  |              return
>  | 
>  | I think ++ is unwanted
>  | 
>  |  new_count = state + 10;
>  |  for (i = fa->state_count; i < new_count; ++i) {
>  |                            ^^^^^
>  |     ...
>  |  }
>  |  fa->state_count = new_count;
>  
>  The meaning of state_count has changed from being the largest number
>  allocated to be the count of the states allocated, so both are correct.
>  The code now passes all the regression tests. We could change the
>  ++state, if we change the resize_state calls to be 1 greater than now,
>  but that is more intrusive.

You are right, both are correct.

Suppose
1) we are here
    918         /* add tmpset to current set of states */
    919         ++(f->curstat);
+   920         resize_state(f, f->curstat);
    921         for (i = 0; i < NCHARS; i++)
2) 100 states was already created and all tables was properly initialized, i.e.
   f->state_count == 2.
3) f->curstat == 99

1) + 2) + 3) ===> unnecessary memory allocation in resize_state(), just a few kbytes.
  Is there any reason for this?

-- 
Best regards, Aleksey Cheusov.